Stereo reconstruction with mixed pixels using adaptive over-segmentation

摘要: We present an over-segmentation based, dense stereo algorithm that jointly estimates segmentation and depth. For mixed pixels on segment boundaries, the computes foreground opacity (alpha), as well color depth for background. model scene a collection of fronto-parallel planar segments in reference view, use generative image formation handles at boundaries. Our method iteratively updates based color, shape constraints using MAP estimation. Given segmentation, are updated belief propagation. show our is competitive with state-of-the-art new Middlebury evaluation, it overcomes limitations traditional methods while properly handling pixels. Z-keying results advantages combining